MKL-DNN DNNL AMD EPYC 2 x AMD EPYC 7601 32-Core testing with a Dell 02MJ3T (1.2.5 BIOS) and llvmpipe 504GB on Ubuntu 19.10 via the Phoronix Test Suite. EPYC 7742 2P: Processor: 2 x AMD EPYC 7742 64-Core @ 2.25GHz (128 Cores / 256 Threads), Motherboard: AMD DAYTONA_X (RDY1001C BIOS), Chipset: AMD Starship/Matisse, Memory: 516096MB, Disk: 280GB INTEL SSDPED1D280GA + 256GB Micron_1100_MTFD, Graphics: llvmpipe 504GB, Monitor: VE228, Network: 2 x Mellanox MT27710 OS: Ubuntu 19.10, Kernel: 5.3.0-13-generic (x86_64), Desktop: GNOME Shell 3.34.0, Display Server: X Server 1.20.5, Display Driver: modesetting 1.20.5, OpenGL: 3.3 Mesa 19.2.0 (LLVM 9.0 128 bits), Compiler: GCC 9.2.1 20190909, File-System: ext4, Screen Resolution: 1920x1080 Xeon Platinum 8280 2P: Processor: 2 x Intel Xeon Platinum 8280 @ 4.00GHz (56 Cores / 112 Threads), Motherboard: GIGABYTE MD61-SC2-00 v01000100 (T15 BIOS), Chipset: Intel Sky Lake-E DMI3 Registers, Memory: 386048MB, Disk: 280GB INTEL SSDPED1D280GA, Graphics: llvmpipe 377GB, Monitor: VE228, Network: 2 x Intel X722 for 1GbE + 2 x QLogic FastLinQ QL41000 10/25/40/50GbE OS: Ubuntu 19.10, Kernel: 5.3.0-13-generic (x86_64), Desktop: GNOME Shell 3.34.0, Display Server: X Server 1.20.5, Display Driver: modesetting 1.20.5, OpenGL: 3.3 Mesa 19.2.0 (LLVM 9.0 256 bits), Compiler: GCC 9.2.1 20190909, File-System: ext4, Screen Resolution: 1920x1080 EPYC 7601 2P: Processor: 2 x AMD EPYC 7601 32-Core (64 Cores / 128 Threads), Motherboard: Dell 02MJ3T (1.2.5 BIOS), Chipset: AMD 17h, Memory: 516096MB, Disk: 280GB INTEL SSDPED1D280GA + 12 x 500GB Samsung SSD 860 + 120GB SSDSCKJB120G7R, Graphics: llvmpipe 504GB, Monitor: VE228, Network: 2 x Broadcom BCM57416 NetXtreme-E Dual-Media 10G RDMA + 2 x Broadcom NetXtreme BCM5720 2-port PCIe OS: Ubuntu 19.10, Kernel: 5.3.0-13-generic (x86_64), Desktop: GNOME Shell 3.34.0, Display Server: X Server 1.20.5, Display Driver: modesetting 1.20.5, OpenGL: 3.3 Mesa 19.2.0 (LLVM 9.0 128 bits), Compiler: GCC 9.2.1 20190909, File-System: ext4, Screen Resolution: 1600x1200 MKL-DNN DNNL 1.1 Harness: IP Batch 1D - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 2.11 |===================== Xeon Platinum 8280 2P . 1.42 |============== EPYC 7601 2P .......... 4.90 |================================================= MKL-DNN DNNL 1.1 Harness: IP Batch All - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 11.10 |=========== Xeon Platinum 8280 2P . 9.83 |========== EPYC 7601 2P .......... 47.83 |================================================ MKL-DNN DNNL 1.1 Harness: IP Batch 1D - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 29.21 |=================================== Xeon Platinum 8280 2P . 2.18 |=== EPYC 7601 2P .......... 40.53 |================================================ MKL-DNN DNNL 1.1 Harness: IP Batch All - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 176.32 |=============================== Xeon Platinum 8280 2P . 2.90 |= EPYC 7601 2P .......... 270.54 |=============================================== MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_3d - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 4.32 |============ Xeon Platinum 8280 2P . 4.70 |============= EPYC 7601 2P .......... 17.27 |================================================ MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_all - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 403.50 |========= Xeon Platinum 8280 2P . 494.94 |=========== EPYC 7601 2P .......... 2040.28 |============================================== MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_3d - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 1391.48 |===================== Xeon Platinum 8280 2P . 2989.04 |============================================== EPYC 7601 2P .......... 2386.59 |===================================== MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_1d - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 2.89 |============================== Xeon Platinum 8280 2P . 1.24 |============= EPYC 7601 2P .......... 4.72 |================================================= MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_3d - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 3.25 |================= Xeon Platinum 8280 2P . 1.15 |====== EPYC 7601 2P .......... 9.37 |================================================= MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_alexnet - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 47.43 |============ Xeon Platinum 8280 2P . 49.79 |============ EPYC 7601 2P .......... 192.49 |=============================================== MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_all - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 9100.13 |======================== Xeon Platinum 8280 2P . 1521.72 |==== EPYC 7601 2P .......... 16846.88 |============================================= MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_all - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 837.06 |=========== Xeon Platinum 8280 2P . 877.84 |============ EPYC 7601 2P .......... 3470.77 |============================================== MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_1d - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 517.95 |=========================== Xeon Platinum 8280 2P . 0.44 | EPYC 7601 2P .......... 900.06 |=============================================== MKL-DNN DNNL 1.1 Harness: Deconvolution Batch deconv_3d - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 846.19 |===================== Xeon Platinum 8280 2P . 1881.87 |============================================== EPYC 7601 2P .......... 1526.11 |===================================== MKL-DNN DNNL 1.1 Harness: Recurrent Neural Network Training - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 814.70 |=============================================== Xeon Platinum 8280 2P . 223.03 |============= EPYC 7601 2P .......... 709.12 |========================================= MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_alexnet - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 540.52 |===================== Xeon Platinum 8280 2P . 19.51 |= EPYC 7601 2P .......... 1179.54 |============================================== MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_googlenet_v3 - Data Type: f32 ms < Lower Is Better EPYC 7742 2P .......... 29.38 |============ Xeon Platinum 8280 2P . 23.08 |========= EPYC 7601 2P .......... 116.50 |=============================================== MKL-DNN DNNL 1.1 Harness: Convolution Batch conv_googlenet_v3 - Data Type: u8s8f32 ms < Lower Is Better EPYC 7742 2P .......... 1580.49 |=========================================== Xeon Platinum 8280 2P . 7.68 | EPYC 7601 2P .......... 1673.36 |==============================================